A Project Foundation Report On

Hexapod-Based Active Stabilization System

for Subsea Sensor Payloads
For the Partial Fulfillment of

Bachelor of Technology
In Department of Radio Physics & Electronics

Name: Saranya Mukherjee

Roll No: T91/ECE/224083

Registration: D01-1111-0051-22

Under the Supervision of

Dr. Anjan Kumar Kundu

Institute of Radio Physics and Electronics

University College of Science and Technology

University of Calcutta
92, A.P.C ROAD, KOLKATA-700009
 CERTIFICATE

TO WHOM IT MAY CONCERN

This is to certify that the foundation of project report entitled

“Hexapod-Based Active Stabilization System for Subsea Sensor
Payloads” is a Bonafide work done under the supervision of Dr.
Anjan Kumar Kundu, Institute of Radio Physics and
Electronics ,by Saranya Mukherjee(Registration no.-
D01-1111-0051-22), final year student of Institute of
Radio Physics and Electronics, University of Calcutta as part of
his 7th semester course work.

Signature of Supervisor

Dr. Anjan Kumar Kundu

INSTITUTE OF RADIOPHYSICS AND
ELECTRONICS UNIVERSITY OF CALCUTTA,
KOLKATA -700009
DATED: 19/12/2025
 DECLARATION BY GROUP MEMBERS

We the group members solemnly declare that the project is based on our own
ideas under the supervision of our guide Dr. Anjan Kumar Kundu. We assert
the statements made and conclusions drawn are an outcome of our project work.
We further certify that:

• The work contained in the report is original and has been done by us under
the general supervision of our guide.
• Whenever we have used materials (text, diagrams, etc.) from other sources,
we have given due credit to them in the text of the report and giving their
details in references.

Saranya Mukherjee
 ACKNOWLEDGEMENT

I am deeply grateful to my guide, Dr. Anjan Kumar Kundu, for his exceptional guidance and
support throughout the course of my work. Under his mentorship, I have the opportunity to
engage with advanced research problems and gain valuable insights into practical scientific
methodologies. He was consistently approachable, patient, and encouraging, offering
constructive feedback and fostering an environment of learning and intellectual growth. His
expertise and commitment greatly enriched my research experience, for which I remain
sincerely thankful. I am also grateful to the Department of Radio physics and Electronics,
University of Calcutta to give me this opportunity to engage in research and allow us to use all
the resources available in the department.
Contents
List of Figures ii

List of Tables ii

1 Introduction 1

2 Stewart Platform 1

3 Literature Review 2

4 Major issues in Stewart platform implementation 3

5 Stewart platform IK 3
5.1 Define coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
5.2 Pose Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
5.3 Rotation Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
5.4 Inverse Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

6 Mechanical design of each actuator 5

6.1 Basic framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
6.2 IK equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
6.3 FK equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
6.4 Dynamical analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
6.5 Static analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
6.6 Overall analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

7 Control system for actuator 10

7.1 Define the macro-variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
7.2 Define the invariant manifold . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
7.3 Formulate the system dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 12
7.4 Formulate the control law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
7.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

8 Effect of SCT on platform’s overall dynamics 14

8.1 Jacobian Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8.2 Jacobian inversion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8.3 Least-Squares Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8.4 Integrating SCT dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
8.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
8.5.1 Case study 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8.5.2 Case study 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
8.5.3 Case study 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

9 Conclusion 20

10 References 20

i
List of Figures
1 Schematic of a Stewart Platform showing the fixed base, moving platform, and
six actuators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Schematic of the proposed actuator design using a scissor mechanism driven by
a rotary servo motor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Geometry of the scissor mechanism showing each term. . . . . . . . . . . . . . 7
4 Free body diagram of the proposed mechanism. . . . . . . . . . . . . . . . . . 8
5 System parameters for actuator with 𝐿 = 8.5𝑐𝑚, 𝑥 = 1.5𝑐𝑚 . . . . . . . . . . . 9
6 Full scale movement vs time for SCT dynamics . . . . . . . . . . . . . . . . . 13
7 Platform motion vs time for one direction of twist . . . . . . . . . . . . . . . . 17
8 Platform motion vs time for three direction of twists . . . . . . . . . . . . . . . 18
9 Platform motion vs time for three direction of twists with varying 𝑇 . . . . . . 19

List of Tables
1 Representative References for Major Control Schemes . . . . . . . . . . . . . . 11

ii
 Hexapod-Based Active Stabilization System for
Subsea Sensor Payloads
Saranya Mukherjee, AK Kundu
IRPE, University of Calcutta

1 Introduction
A Stewart platform, also known as a hexapod, is a type of parallel manipulator defined by
a moving platform connected to a fixed base by six linear actuators (legs). This parallel
structure grants the platform six degrees of freedom (6-DOF)—three translational and three
rotational—while simultaneously providing high stiffness, exceptional load-carrying capacity,
and precise kinematic control.
These characteristics make it ideal for tasks requiring high dynamic accuracy and rapid
response, leading to uses in flight simulators, high-precision machining, and surgical systems.
In the context of subsea robotics, the Stewart platform functions as an active stabilization end-
effector for critical sensor payloads (like cameras or sonars). Its 6-DOF capability allows it to
actively and instantaneously compensate for the base vehicle’s motion (pitch, roll, and surge) and
the unevenness of the seabed. This compensation ensures that the sensor maintains a constant,
precise standoff distance and orientation relative to the target, which is essential for maximizing
data quality during bathymetric surveys and targeted visual inspection. Furthermore, it allows for
fine manipulation and micro-adjustments over specific seabed features, effectively decoupling
the sensor’s motion from the movement of the larger autonomous or remotely operated vehicle.

2 Stewart Platform
A Stewart Platform, formally known as a parallel manipulator or hexapod, is a mechanical
device that represents a significant departure from conventional serial-link robots. Unlike an
articulated arm (serial robot) where one joint follows another, the Stewart platform consists
of a moving platform connected to a fixed base via six independent and synchronized linear
actuators (legs). The key characteristic is that all six legs work in parallel to control the motion
of the platform. This parallel arrangement is what grants the platform its immense mechanical
advantages; the load and forces are distributed across multiple supports, it achieves exceptionally
high stiffness, better precision, and superior dynamic response compared to serial manipulators
of similar size. This architecture inherently restricts the accumulation of positioning errors,
leading to high repeatability.
The power of the Stewart platform lies in its ability to control all six degrees of freedom
(6-DOF) simultaneously: three translational motions (surge, sway, and heave—movement along
X, Y, and Z axes) and three rotational motions (roll, pitch, and yaw). Control is achieved by
precisely varying the length of each of the six actuators. To move the platform in any single
direction or rotation, a coordinated change in the length of multiple actuators is required. This

1
coordination is computationally intensive, requiring a sophisticated inverse kinematic model to
translate the desired position and orientation of the top platform into the specific lengths for the
six individual actuators.
The unique combination of high dynamic performance and stability makes Stewart plat-
forms indispensable in critical applications. Their original development was for use in aircraft
and driving simulators where they must accurately replicate vehicle motion and forces instan-
taneously across all six axes to provide a realistic experience. In industrial settings, they are
employed as high-speed positioners for tasks like laser welding, medical imaging, and ortho-
pedic surgery positioning, where sub-millimeter accuracy is crucial. For subsea robotics, the
parallel structure is invaluable for active vibration isolation and stabilization. By continuously
adjusting the six legs in real-time based on sensor feedback (like altimeters or gyroscopes), the
platform can effectively nullify the movement of the larger vehicle chassis, ensuring that the
sensor payload remains perfectly level and at a precise distance from the seabed regardless of
wave action or vehicle turbulence. This ability to instantly compensate for external disturbances
is the core reason for their utility in dynamic and hostile environments.

3 Literature Review
We can split the existing literature on Stewart platforms into two major categories:
• Kinematic Modeling and Trajectory Planning: These involve understanding the platform’s
geometry and motion. Studies have documented various configurations, such as the 6-6 [7]
and 6-3 [10] designs, which are defined by the spacing and arrangement of the joints on
the fixed and mobile platforms. On the practical side, research has detailed the analysis of
the platform’s design, kinematic modeling, and trajectory generation. This work includes
the application of inverse kinematics to define specific movements, such as point-to-point
transfers via cycloidal trajectories [16] and more complex motions simulating oceanic
waves. Furthermore, the computational efficiency of direct kinematics is studied to
enable real-time feedback and allow for advanced control loops, such as ”tracking error”
supervisors.
• Control Strategies and Algorithms:
Sliding Mode Control (SMC): SMC and its derivatives are popular for their robust-
ness. A new controller based on the adaptive sliding-mode technique and a ”Super-
Twist” [8] state observer was proposed to manage nonlinear systems and external per-
turbations, demonstrating system stability. Similarly, other research introduced a novel
adaptive super-twisting based smooth integral sliding mode controller (ASTSISMC) [15],
where the super-twisting control replaces the discontinuous part of the integral sliding
mode, and its gains adapt to the amplitude of disturbances, significantly improving dis-
turbance rejection. A more generalized SMC [17] was also proposed and experimentally
validated to improve the performance and positioning accuracy of Stewart platforms in
industrial domains.
Model Predictive Control (MPC): For dynamic trajectory following, a Switchable
Model Predictive [20] Control (SMPC) strategy has been developed, focusing on repli-
cating motion feelings during pilot training in flight simulators. Additionally, Iterative
Learning Control (ILC) [9] has been used to accurately replicate a desired vibration
response for road test simulation systems, demonstrating its utility in product quality
enhancement and development time reduction for vehicle manufacturers.

2
 Active Disturbance Rejection Control (ADRC): To combat issues like tracking lag
from overload [13] and the impact of nonlinear factors, an improved Active Disturbance
Rejection Control (ADRC) strategy for hydraulic-driven platforms was proposed. This
method utilizes a cascaded extended state observer (ESO) to enhance the system’s anti-
interference ability and provide faster response times than traditional control methods.

4 Major issues in Stewart platform implementation

While Stewart platforms offer significant advantages, several challenges must be addressed for
effective implementation:
• Complexity of Kinematics: The inverse kinematics of Stewart platforms is inherently
complex due to the parallel structure and the interdependence of the six actuators. This
complexity can lead to computational challenges, especially for real-time applications
where rapid response is required.
• Control System Design: Designing robust control systems that can handle the nonlinear
dynamics and potential disturbances in real-world applications is challenging. Advanced
control strategies, such as sliding mode control or model predictive control, are often
necessary but can be difficult to implement effectively.
• Mechanical Design and Fabrication: The precision required in the mechanical design and
fabrication of Stewart platforms can be demanding. Any inaccuracies in the construction
can lead to performance issues, such as reduced stiffness or increased backlash.
• Cost and Maintenance: Stewart platforms can be expensive to build and maintain due
to their complexity and the precision required in their components. This can limit their
accessibility for certain applications.
The foundational work aims at addressing the mechanical and cost related challenges through
innovative design, which shall later pave the way for advanced control strategies to be imple-
mented effectively.

5 Stewart platform IK
5.1 Define coordinate systems
To solve the inverse kinematics problem, we define two coordinate systems:
• Fixed Base Coordinate System( 𝐵 ): Origin at the center of the fixed base platform.
Í

• Moving Platform Coordinate System( 𝑃 ): Origin at the center of the moving platform.
Í

Hence, the geometry of the Stewart platform can be defined by:

• 𝐵𝑖 : Position vectors of the base attachment points in 𝐵 .
Í

• 𝑃𝑖 : Position vectors of the platform attachment points in

Í
𝑃

• T: Translational vector of the platform with respect to the base.

• R: Rotation matrix representing the orientation of the platform with respect to the base.

3
Figure 1: Schematic of a Stewart Platform showing the fixed base, moving platform, and six
actuators.

5.2 Pose Representation

The pose of the moving platform is defined by a 6-element vector:
𝑥 
 
𝑦
 
𝑧
p =   (1)
𝛼 
 𝛽
 
𝛾 
 
Where (𝑥, 𝑦, 𝑧) is the translation and (𝛼, 𝛽, 𝛾) represent roll, pitch, and yaw in radians.

5.3 Rotation Matrix

Platform orientation is modeled by the combined rotation matrix:

R = 𝑅𝑧 (𝛾)𝑅 𝑦 (𝛽)𝑅𝑥 (𝛼) (2)

Where:
1 0 0   cos 𝛽 0 sin 𝛽 
  
𝑅𝑥 (𝛼) = 0 cos 𝛼 − sin 𝛼
 𝑅 𝑦 (𝛽) =  0 1 0 
0 sin 𝛼 cos 𝛼  − sin 𝛽 0 cos 𝛽
   
cos 𝛾 − sin 𝛾 0
 
𝑅𝑧 (𝛾) =  sin 𝛾 cos 𝛾 0 (3)
 0 0 1


4
5.4 Inverse Kinematics
For each actuator: Given the base point b𝑖 and platform point p𝑖 , The transformed platform
joint position is:
P𝑖𝑤𝑜𝑟𝑙𝑑 = R · P𝑖𝑙𝑜𝑐𝑎𝑙 + T (4)
Where, P𝑖𝑙𝑜𝑐𝑎𝑙 is the pose in terms of the 6-element vector, and P𝑖𝑤𝑜𝑟𝑙𝑑 is its transformation.
So the vector along the actuator becomes:

L𝑖 = P𝑖𝑤𝑜𝑟𝑙𝑑 − B𝑖 (5)

So the actual length of the actuator is:

√︃
𝑤𝑜𝑟𝑙𝑑 − 𝐵 ) 2 + (𝑃 𝑤𝑜𝑟𝑙𝑑 − 𝐵 ) 2 + (𝑃 𝑤𝑜𝑟𝑙𝑑 − 𝐵 ) 2
𝐿 𝑖 = ||L𝑖 || = (𝑃𝑖𝑥 (6)
𝑖𝑥 𝑖𝑦 𝑖𝑦 𝑖𝑧 𝑖𝑧

As is evident from the above equations, by substituting the desired pose parameters into
the transformation equations, we can compute the required lengths of each of the six actuators
to achieve that pose. But this process is computationally intensive due to the trigonometric
functions involved in the rotation matrix. This specific issue can mostly be addressed by
moderately sized lookup tables and interpolation techniques for real-time applications, aided
by modern computational power and specifically designed hardware in embedded controllers.

6 Mechanical design of each actuator

Traditonal linear actuators used in Stewart platforms include hydraulic cylinders, electric linear
actuators and alike but all of these options are expensive and complex to maintain in subsea
environments. Not only this, but also either precision or speed is compromised in these
traditional actuators. We have tried to come up with a novel mechanical design for each actuator
that uses a combination of a common DC servos, and a scissor mechanism to convert rotary
motion to linear motion while balancing not only both high precision and speed and cost-
effectiveness, but also ease of maintenance in subsea environments and torque requirements
within the limits of common hobby servos.

6.1 Basic framework

The basic framework of the proposed actuator design is shown in Fig. 2. The scissor mechanism
consists of four arms of length 𝐿 each, connected at a pivot point by two longer arms of length
2𝐿. The bottom ends of the shorter arms are connected to a fixed base, while the top ends are
connected to the moving platform. The longer arms are connected to a servo motor at the pivot
point, which drives the rotation of the mechanism. As the servo rotates, it changes the angle 𝛽
between the longer arms and the horizontal plane. This rotation causes the scissor mechanism
to extend or retract, changing the vertical height ℎ of the moving platform. The relationship
between the angle 𝛽 and the height ℎ can be derived using trigonometric principles, as discussed
in the next section. The entire mechanism is based upon the assumption that the motion system
stays symetrical about the vertical axis (marked in red). The mechanisms to enfore the later
shall be discussed in the detailed design section.

5
Figure 2: Schematic of the proposed actuator design using a scissor mechanism driven by a
rotary servo motor.

6.2 IK equations
Fig. 3 shows the geometry of the scissor mechanism. From the figure, we shall focus on the
triangle ABC. Here,
𝐻 = 2ℎ
𝛽 = 2𝛼
𝑥 
𝜃 = tan−1
ℎ
𝜋
𝛾 = − (𝛼 + 𝜃)
2
√︁
𝑑 = 𝑥 2 + ℎ2
Using the cosine rule in triangle ABC, we get:
𝐿2 + 𝑑2 − 𝐿2 𝑑
cos 𝛾 = =
2𝐿𝑑 2𝐿
Given that cos 𝛾 = cos( 𝜋2 − 𝛼 − 𝜃) = sin(𝛼 + 𝜃), we have:
   
−1 𝑥 −1 𝑑
𝛼 + tan = sin
ℎ 2𝐿
From this, we can solve for 𝛼:
√ !
𝑥 2 + ℎ2  
−1 𝑥
𝛼 = sin−1 − tan
2𝐿 ℎ

6
 Figure 3: Geometry of the scissor mechanism showing each term.

Finally, we can compute 𝛽 as:

√︃
𝐻2
 2   
 −1 ©­ 𝑥 + ® − tan−1 2𝑥 

4 ª
𝛽 = 2 sin ­ (7)
 2𝐿 ® 𝐻 
 
 « ¬ 
Since, all other parameters are in degrees,
180
𝛽𝑑𝑒𝑔 = 𝛽 · (8)
𝜋
Equation 7 gives us the inverse kinematics equation for the scissor mechanism, which shall
serve as the basis for the control of each actuator in the Stewart platform.
IK equations are required to compute the required servo angle 𝛽 for a desired actuator length
𝐻, but FK equations are also necessary inroder to formulate the control law and compute the
dynamics of the motion of the systems. So the next section discusses the FK equations of the
scissor mechanism.

7
 Figure 4: Free body diagram of the proposed mechanism.

6.3 FK equations
From Fig. 3, we can use the sine rule in triangle ABC to get:
√︂
(𝐿 cos 𝛼 − 𝑥) 2
ℎ = 𝐿 sin 𝛼 + 𝐿 1 −
𝐿2
Hence, we can compute 𝐻 as:
√︄
(𝐿 cos 2𝛽 − 𝑥) 2 
 
 𝛽
𝐻 = 2𝐿 sin + 1 −  (9)
 2 𝐿2 

 
Now that we have both IK and FK equations for the scissor mechanism, we can proceed to
the dynamic analysis for this mechanism.

6.4 Dynamical analysis

To derive the dynamics of the scissor mechanism, we can use differenciate the FK and IK
equations with respect to time.
Differentiating Equation 9 with respect to time gives us the velocity relationship:
 
   
𝑑𝐻 𝑑𝛽  𝛽 1 𝛽 𝛽 
= · cos + · sin cos − (10)

 𝐿 √︂ 𝐿 𝑥 
𝑑𝑡 𝑑𝑡  2 𝛽
(𝐿 cos 2 −𝑥) 2 2 2 

 1− 𝐿 2


 
Similarly, differentiating Equation 7 with respect to time gives us:

© ª
𝑑𝛽 ­­ 𝐻 1 4𝑥 ®® 𝑑𝐻
= · √︂ + 2 (11)
𝑑𝑡 ­­ 4𝐿 𝑥 2 + 𝐻4
2 √︃ 4𝑥 + 𝐻 2 ®® 𝑑𝑡
𝐻2
1− 4𝐿 2
· 𝑥2 + 4
« ¬

8
 (a) Actuator length 𝐻 vs 𝛽 (b) Torque on Servo 𝜏 vs 𝐻

(c) dH/dt vs 𝛽 for varying 𝑑𝛽/𝑑𝑡

Figure 5: System parameters for actuator with 𝐿 = 8.5𝑐𝑚, 𝑥 = 1.5𝑐𝑚

6.5 Static analysis

We need to compute the torque required at the servo motor to hold the platform at a certain
height 𝐻. Fig. 4 shows the free body diagram of the scissor mechanism. Assuming that a force
𝐹 acts at the center of the platform, since the mechanism is symmetrical about the vertical axis,
each half will experience a force of 𝐹/2.
Balancing forces at the top platform link,
𝐹
𝐹1 cos(𝛾 − 𝜃) =
2
𝜙 = 𝜋 − 2𝛾

9
Forces acting on the middle link will be 𝐹1 on both sides for the link to be at equilibrium. So,

𝐹𝑡 = 𝐹1 sin 𝜙 = 𝐹1 sin(𝜋 − 2𝛾) = 𝐹1 sin(2𝛾) = 2𝐹1 sin 𝛾 cos 𝛾

𝐹 sin 𝛾 cos 𝛾
𝐹𝑡 =
𝑐𝑜𝑠(𝛾 − 𝜃)
Now, given that the link has a length 𝐿 and the torque is distributed evenly on both sides,

𝐹 · 𝐿 sin (2 ∗ (𝜃 + 𝛼)) 2𝐿 · 𝐹 sin(2𝛾)

𝜏= = (12)
sin (2𝜃 + 𝛼) sin 𝛼
    √︃
−1 2𝑥 −1 2
Where, 𝜃 = tan 𝐻 , 𝛾 = cos
𝑑
2𝐿 , 𝑎𝑛𝑑 𝑑 = 𝑥 2 + 𝐻4 .

6.6 Overall analysis

The following points summarize the performance analysis of the proposed actuator, drawing
conclusions from the height, torque, and velocity curves(Fig. 5).

• Height-Angular Position Curve(Fig. 5a): The mechanism exhibits a large and highly linear
region of operation, which is extremely desirable for simple and predictable control. The
system provides a significant range of motion(ROM), extending from approximately 10𝑐𝑚
to about 33.8𝑐𝑚.

• Torque-Height Curve(Fig. 5b): Analyzing the torque requirements, the focus is placed on
the region where the required torque falls below the typical 10𝑁 · 𝑐𝑚 holding torque of
standard commodity servos. Based on the curve for a 300𝑔𝑚 force, the actuator provides
an actually usable range of motion from approximately 21.3𝑐𝑚 to 33.8𝑐𝑚, which falls
within the feasible operational zone of low-cost servos.

• Linear Velocity-Angular Velocity Curve(Fig. 5c): This study confirms the actuator’s
feasibility for operation in linear velocity mode in addition to linear position mode.
The graph demonstrates that the mechanism can achieve a very high linear velocity for
relatively low angular velocities of the servo motor. Notably, the highest point recorded
shows a linear velocity of 30.6𝑚𝑚/𝑠 is achievable with an angular speed as low as
0.1𝑟𝑎𝑑/𝑠𝑒𝑐 (approximately 5.7 degrees per second).

These results conclusively demonstrate that our proposed actuator not only provides a massive
cost advantage over traditional linear actuators but also matches or surpasses them in every
critical performance benchmark, including range of motion, control linearity, and attainable
linear velocity.

7 Control system for actuator

Synergetic control theory (SCT) is a nonlinear control framework that shapes system behav-
ior by forcing its states toward an invariant manifold, where the closed-loop dynamics [1]
evolve smoothly and predictably. Unlike switching-based methods such as sliding-mode con-
trol [4], SCT [11] generates a continuous control law, eliminating chattering and yielding
gentler actuator effort. Compared with classical linear approaches—like PID or state-feedback

10
 Control Scheme Reference Key Contribution / Notes
Synergetic Control Theory [4, 11] Demonstrates chattering-free dynamics
(SCT) vs. SMC and improved smooth-
ness/robustness.
Sliding-Mode Control (SMC) [8, 20] Surveys robustness and finite-time con-
vergence; discusses chattering issues.
Classical / Linear Controllers [13, 15] Evaluates classical PID limitations for
(PID, LQR, State Feedback) nonlinear and uncertain systems.
Fuzzy Logic Controllers [3, 19] Comprehensive survey of FLC and
(FLC) neuro-fuzzy control methods.
Neural-Network / Learning- [2] Overview of NN- and FLC-based con-
Based Control trol, stability issues, and design methods.
Adaptive / Backstepping / [2, 14] Survey of adaptive, robust, and con-
Modern Nonlinear Control strained backstepping methods.

Table 1: Representative References for Major Control Schemes

control—synergetic [14] control offers stronger robustness to parameter variations and non-
linearities because its structure explicitly embeds system dynamics into a coordinated macro-
variable [1].
Table 1 summarizes key references for various control schemes applied to Stewart platform
like nonlinear systems. While PID controllers aim to regulate individual error signals indepen-
dently, SCT coordinates multiple state variables toward a unified goal, improving performance
in multi-state nonlinear systems. Relative to more computationally intensive intelligent con-
trollers [2, 3, 19] (e.g., fuzzy logic or neural networks), SCT retains analytical transparency
and simpler real-time implementation while still delivering high stability and fast conver-
gence [1]. Overall, synergetic control theory provides a powerful middle ground: more robust
and nonlinear-aware than classical controllers, smoother and easier to digitize than sliding-mode
control, and more interpretable and lightweight than many AI-based schemes.
Moving forward, we plan to develop a synergetic control law for the Stewart platform using
the derived inverse kinematics and dynamic model.

7.1 Define the macro-variable

To design a synergetic controller, we first define a macro-variable that captures the desired
system behavior. Currently, we are looking for a effective position hold actuator that can reach
the desired length without overshoot or oscillations. For the Stewart platform, we can define
the macro-variable
𝜎 = 𝐻𝑓 − 𝐻 (13)
Where 𝐻 𝑓 is the desired actuator length and 𝐻 is the current length.

11
7.2 Define the invariant manifold
Next, we define the invariant manifold that the system should converge to. A common choice is
to set a first degree differencial equation for 𝜃:

𝑇 ·𝜎
¤ +𝜃 =0 (14)

Where, 𝜃 = 𝑓 (𝜎) is a nonlinear function of 𝜎 that defines the desired dynamics on the manifold,
and 𝑇 is a designer chosen fixed parameter, that determines the convergence speed (smaller 𝑇
means faster convergence). 𝜃 can be chosen as:
• invertible and differenciable
• 𝜃 (0) = 0
• 𝜎 · 𝜃 (𝜎) > 0 for all 𝜎 ≠ 0
We have chosen:
𝜃=𝜎

7.3 Formulate the system dynamics

From eqn. 14,  
𝑑𝐻 𝑓 𝑑𝐻  
𝑇· − + 𝐻𝑓 − 𝐻 = 0 (15)
𝑑𝑡 𝑑𝑡
Since, 𝐻 𝑓 is constant for position hold,
 
𝑑𝐻  
𝑇· − + 𝐻𝑓 − 𝐻 = 0
𝑑𝑡
Now, substituting eqn. 9 and 10,
  
    
 𝑑𝛽  𝛽 1 𝛽 𝛽 
𝑇 · −  𝐿 · cos + √︂ · sin 𝐿 cos − 𝑥  
  
 𝑑𝑡  2 (𝐿 cos 𝛽2 −𝑥) 2
2 2 



 1 − 𝐿 2


  √︄  
(𝐿 cos 2𝛽 − 𝑥) 2  
  
  𝛽
+ 𝐻 𝑓 − 2𝐿 sin + 1 −
 
2
  = 0 (16)
  2 𝐿 

  
𝑑𝛽
Now, solving for 𝑑𝑡 ,

𝑑𝛽
=
𝑑𝑡 √︄
(𝐿 cos 2𝛽 − 𝑥) 2  
  
1   𝛽
· 𝐻 𝑓 − 2𝐿 sin + 1− 

    
 2
 𝐿2 

1
𝑇 ·  𝐿 · cos 2𝛽 + √︂
𝛽
· sin 𝛽
2 𝐿 cos 2𝛽 − 𝑥    
(𝐿 cos − 𝑥 ) 2
 1− 2 
 𝐿2 
(17)

12
 (a) Actuator length 𝐻 vs Time

(b) Servo Angle 𝛽 vs Time

Figure 6: Full scale movement vs time for SCT dynamics

7.4 Formulate the control law

Finally, we use eqn. 17 to get the control law:
¤
max | 𝛽(𝑡)| = 250◦ /𝑠 = 4.36𝑟𝑎𝑑/𝑠
¥
max | 𝛽(𝑡)| = 4583.66◦ /𝑠2 = 80𝑟𝑎𝑑/𝑠2
We need to enfore these hard conditions onto the system dynamics, to satisfy the maximum
limits of a realistic comodity servo motor.
∫
𝑑𝛽
𝛽= 𝑑𝑡 (18)
𝑑𝑡
Now descretizing the equation for digital implementation, we get:
𝑑𝛽
𝛽[𝑘 + 1] = 𝛽[𝑘] + · Δ𝑡 (19)
𝑑𝑡
Where 𝛽[𝑘] is the servo angle at time step 𝑘, and Δ𝑡 is the sampling time, and 𝛽[𝑘 + 1] is the
updated servo angle to be sent to the servo motor.

13
7.5 Results
Fig. 6 shows the simulation results of the synergetic control law for a full scale movement from
𝐻 = 21.3𝑐𝑚 to 𝐻 = 33.8𝑐𝑚. As seen from Fig. 6a, the actuator reaches the desired length
without any overshoot or oscillations, demonstrating the effectiveness of the control law. Fig. 6b
shows the corresponding servo angle 𝛽 over time, which smoothly transitions to the required
angle to achieve the desired actuator length.
With optimization of the parameter 𝑇 = 0.0001, determined by limiting the maximum
angular velocity of the servo motor at 4.36𝑟𝑎𝑑/𝑠 or 250◦ /𝑠, allowing for ample headroom for
real-world implementation,

• Convergence time(2% band): 0.255s

• Rise time: 0.162s - 0.049s(90% - 10%) = 0.113s

• Steady state error: 0cm

• Overshoot: NA

8 Effect of SCT on platform’s overall dynamics

The Stewart platform in its forward kynematic model is a 40 order nonlinear system. Such
system cannot be analysed in a closed form. However, we can compute the overall dynamics
of the platform using the jacobian of the system and the dynamics of each actuator. Let the
jacobian of the Stewart platform be 𝐽. The end of the platform position is given by:

P𝑖𝑤𝑜𝑟𝑙𝑑 = R · P𝑖𝑙𝑜𝑐𝑎𝑙 + T (20)

Where, P𝑖𝑙𝑜𝑐𝑎𝑙 is the pose in terms of the 6-element vector, and P𝑖𝑤𝑜𝑟𝑙𝑑 is its transformation. So
the vector along the actuator becomes:

L𝑖 = R · P𝑖𝑙𝑜𝑐𝑎𝑙 + T − B𝑖 (21)

Differentiating 𝐿 𝑖 with respect to time yields

¤𝑖
L𝑇𝑖 L
𝐿¤ 𝑖 = = L̂𝑇𝑖 L¤ 𝑖 (22)
∥L𝑖 ∥
The derivative of the leg vector is
¤ 𝑖 = T¤ + R
L ¤ · P𝑖 (23)
Substituting into 𝐿¤ 𝑖 :
𝐿¤ 𝑖 = L̂𝑇𝑖 T¤ + L̂𝑇𝑖 (𝜔 × (R · P𝑖 )) (24)
Defining the platform velocity as  
x¤
v= (25)
𝜔
we obtain  
x¤
𝐿¤ 𝑖 = L̂𝑇𝑖
 
L̂𝑇𝑖 [R · P𝑖 ] · (26)
𝜔

14
8.1 Jacobian Matrix
Stacking all six legs yields
L¤ = J · v (27)
where
L¤ = 𝐿¤ 1 𝐿¤ 2 · · · 𝐿¤ 6
 𝑇

 𝑇
v = 𝑥¤ 𝑦¤ 𝑧¤ 𝜔𝑥 𝜔 𝑦 𝜔 𝑧
and the Jacobian is
L̂𝑇 L̂𝑇 [R · P1 ] 
 1 1 
L̂𝑇 L̂𝑇 [R · P2 ] 
J =  ..2 2
(28)
 
 . .
. 
 𝑇 . 

L̂ L̂𝑇 [R · P6 ] 
 6 6 

8.2 Jacobian inversion

¤ we need to invert
Since, we need to compute the platform velocity v from the leg velocities L,
the jacobian matrix J. Assuming that the platform is not in a singular configuration, we can
compute the inverse of the jacobian matrix using the pseudo-inverse [5]. Normally inverting
the jacobian matrix directly is:

• Computationally expensive

• Numerically unstable near singularities

Hence, we choose to switch the problem to a damped least-squares optimization problem [6],
which provides a more robust and stable solution with the assumption that:

• The platform velocity v should be minimized

• The leg velocities L

¤ should be tracked as closely as possible

• Large platform velocities are penalized

8.3 Least-Squares Formulation

The inverse velocity kinematics problem is formulated as the minimization of the squared error
¤ 2
min ∥Jv − L∥ (29)
v

The solution satisfies the normal equations

¤
J𝑇 Jv = J𝑇 L (30)

If J𝑇 J is singular, this system has no unique solution.

To ensure a unique and bounded solution, a regularization term is introduced[Tikhonov
regularization] [12, 18]:  
min ∥Jv − L∥¤ 2 + 𝜆2 ∥v∥ 2 (31)
v
where 𝜆 > 0 is the damping factor.

15
 This formulation penalizes large platform velocities and improves numerical stability.
Expanding the objective function:
¤ 𝑇 (Jv − L)
𝑓 (v) = (Jv − L) ¤ + 𝜆2 v𝑇 v (32)
Taking the gradient with respect to v and setting it to zero:
¤ + 2𝜆2 v = 0
∇ 𝑓 = 2J𝑇 (Jv − L) (33)
Since 𝜆2 I is positive definite, the matrix J𝑇 J + 𝜆2 I is invertible for all 𝜆 > 0. Thus,
v = (J𝑇 J + 𝜆2 I) −1 J𝑇 L¤ (34)
Using the identity
(J𝑇 J + 𝜆2 I) −1 J𝑇 = J𝑇 (JJ𝑇 + 𝜆2 I) −1
the damped least-squares inverse can be written as
v = J𝑇 (JJ𝑇 + 𝜆2 I) −1 L¤ (35)
This form is preferred when 𝑚 ≤ 𝑛, as in Stewart platforms. The damping factor 𝜆:
• prevents unbounded velocities near singular configurations,
• ensures smooth solutions,
• trades tracking accuracy for robustness.
As 𝜆 → 0, the solution approaches the Moore–Penrose pseudoinverse, but as 𝜆 increases,
the solution minimizes ∥v∥ rather than tracking accuracy.

8.4 Integrating SCT dynamics

To integrate the SCT-based actuator dynamics into the overall platform dynamics, we substitute
the leg velocity vector L¤ with the SCT-derived velocities.
Note: L is the length of the actuator, while 𝐿 is the length of the scissor link.
From eqn. 17, we get an expression for 𝑑𝛽𝑑𝑡 :
𝑑𝛽
= 𝑓 (L, 𝐻 𝑓 , 𝐿, 𝑥, 𝑇) (36)
𝑑𝑡
𝑑𝐻 𝑑𝛽
Using eqn. 10, we can compute 𝑑𝑡 as a function of 𝑑𝑡 .
√︄
(𝐿 cos 2𝛽 − 𝑥) 2  
  
¤ 1   1   𝛽
L= 𝐻 𝑓 − 𝐻 = 𝐻 𝑓 − 2𝐿 sin + 1 −
 
2
 (37)
𝑇 𝑇   2 𝐿 

  
When this equation is substituted into the damped least-squares inverse kinematics formu-
lation, we obtain the platform velocity v that respects the SCT-based actuator dynamics. So, we
get,
6
√︄
  𝛽𝑖 2
1 𝛽𝑖 (𝐿 cos 2 − 𝑥)  
v = J𝑇 (JJ𝑇 + 𝜆2 I) −1 · 𝐻 𝑓 − 2𝐿 sin + 1 −
 
2
 (38)
𝑇   2 𝐿 

    𝑖=1
So, the overall platform dynamics now incorporate the SCT-based control of each actuator,
enabling coordinated motion that respects the individual actuator constraints and dynamics. It
can be observed that the platform position will also follow the same nature of the curve as
Fig. 6a, but in 6 degrees of freedom.

16
 (a) Actuator length L vs Time

(b) Platform vertical position 𝑍 vs Time

Figure 7: Platform motion vs time for one direction of twist

8.5 Results
Define the base positions as:
−120.18 120.18 224.80 104.80 −104.80 −224.80
 
B =  190.81 190.81 17.70 −199.67 −199.67 17.70 
 0 0 0 0 0 0 

Define the platform positions as:
−46.71 46.71 100.99 53.56 −53.56 −100.99
 
P =  88.99 88.99 −7.88 −85.04 −85.04 −7.88 
 250 250 250 250 250 250 

All values are in mm. The P positions correspond to an initial actuator length of 𝐻 = 28𝑐𝑚.
Here, the platfrom is initially parallel to the base, thus all actuators have the same length.

17
 (a) Z velocity of the platform J(3) vs Time

(b) Platform pitch vs Time

Figure 8: Platform motion vs time for three direction of twists

8.5.1 Case study 1

Vertical motion of the platform from 𝑧 = 25𝑐𝑚 to 𝑧 = 29.4𝑐𝑚. Only one direction of twist is
considered here for simplicity. We will analyse the platform’s motion with respect to each of
the actuators moving simultaneously. Fig. 7 shows the simulation results of the platform motion
for one direction of twist. It demonstrates the effectiveness of the SCT-based actuator dynamics
on reflecting the same smooth manifold onto the overall platform motion accurately.

8.5.2 Case study 2

Twisting motion of the platform on all axes are considered simultaneously. The platform is
twisted from an initial orientation of [0◦ , 0◦ , 0◦ ] to [−3.5◦ , 11.3◦ , 31.4◦ ], along with a vertical
displacement from 𝑧 = 25𝑐𝑚 to 𝑧 = 26.1𝑐𝑚, with minimal twist in the translational X and Y.
Here we shall analyse the all the twist directions of the platform and their dynamics. Fig. 8 shows
the simulation results of the platform motion for three directions of twist. We need to make sure

18
 (a) Z velocity of the platform J(3) vs Time when 𝑇 is too high

(b) Z velocity of the platform J(3) vs Time when 𝑇 is too low

Figure 9: Platform motion vs time for three direction of twists with varying 𝑇

that the velocity of the platform(J) is smooth and convergent throughout the motion(Fig. 8a).
In order to keep J smooth, we derived an emperical relation,
𝑑𝑡
< 0.2 (39)
𝑇
Where, 𝑑𝑡 is the computation time step, and 𝑇 is the SCT parameter.

8.5.3 Case study 3

As a continuation of case study 2, we shall analyse the platform’s velocity curves with respect
to time. We shall analyse the rate of change of velocites with respect to controller parameter
𝑇. We shall also optimize the parameter 𝑇 to achieve the fastest convergence while maintaing
smooth motion throughout, on all degrees of freedom. Fig. 9 shows the simulation results of the
platform motion for three directions of twist with varying 𝑇. It demonstrates the effect of the
SCT parameter 𝑇 on the platform’s overall dynamics. When 𝑇 is too high(Fig. 9a), the platform

19
velocity J(3) is very jumpy and oscillatory, which is undesirable for smooth motion. When 𝑇 is
too low(Fig. 9b), the platform velocity J(3) is smooth but very slow to converge, which is also
undesirable for fast response. Hence, we need to optimize 𝑇 such that the platform velocity J,
to get the optimal response as in Fig. 8a.

9 Conclusion
The work presented demonstrates the feasibility of a novel, low-cost scissor-mechanism actuator
as a viable substitute for traditional linear actuators within a Stewart platform. Through the
development of complete inverse and forward kinematic models, followed by detailed static,
dynamic, and performance analyses, the proposed design shows excellent linearity, a wide usable
range of motion, and actuator torque requirements that fit comfortably within the capabilities
of standard hobby-grade servos.
Simulation results further confirm that the mechanism can achieve high linear velocities
and maintain precise positioning despite its mechanical simplicity. Building on this mechanical
foundation, a synergetic control law was formulated using the system’s derived dynamics. The
controller successfully drives the actuator to its target position without overshoot, oscillation, or
chattering, achieving fast convergence and zero steady-state error. Together, the actuator design
and the SCT-based control approach form a strong basis for implementing a complete, actively
stabilized Stewart platform suitable for subsea sensing applications. Future work will integrate
the actuator modules into the full hexapod structure and extend the synergetic controller to
coordinate all six degrees of freedom for real-time platform stabilization.

10 References
[1] Azadeh Ahifar, Abolfazl Ranjbar Noee, and Zahra Rahmani. Terminal synergetic design
of a nonlinear robot manipulator in the presence of disturbances. COMPEL-The interna-
tional journal for computation and mathematics in electrical and electronic engineering,
37(1):208–223, 2018.
[2] Mojtaba Hadi Barhaghtalab, Mohammadreza Askari Sepestanaki, Saleh Mobayen, Abol-
fazl Jalilvand, Afef Fekih, and Vahid Meigoli. Design of an adaptive fuzzy-neural infer-
ence system-based control approach for robotic manipulators. Applied Soft Computing,
149:110970, 2023.
[3] Zoulikha Bouhamatou and Foudil Abedssemed. Fuzzy synergetic control for dynamic
car-like mobile robot. Acta Mechanica et Automatica, 16(1):48–57, 2022.
[4] Nourdine Bounasla, Kamel Eddine Hemsas, and Hacene Mellah. Synergetic and sliding
mode controls of a pmsm: A comparative study. Journal of Electrical and Electronic
Engineering, 3(1-1):22–26, 2015.
[5] Samuel R Buss. Introduction to inverse kinematics with jacobian transpose, pseudoinverse
and damped least squares methods. IEEE Journal of Robotics and Automation, 17(1-
19):16, 2004.
[6] S. Chiaverini. Singularity-robust task-priority redundancy resolution for real-time kine-
matic control of robot manipulators. IEEE Transactions on Robotics and Automation,
13(3):398–410, 1997.

20
 [7] Bhaskar Dasgupta and TS1739334 Mruthyunjaya. The stewart platform manipulator: a
review. Mechanism and machine theory, 35(1):15–40, 2000.

[8] Sajjad Keshtkar, Alexander S Poznyak, E Hernandez, and Armando Oropeza. Adaptive
sliding-mode controller based on the “super-twist” state observer for control of the stewart
platform. Automation and Remote Control, 78(7):1218–1233, 2017.

[9] Byeongsik Ko, Jong-Wook Park, and Dong W Kim. A study on iterative learning control for
vibration of stewart platform. International Journal of Control, Automation and Systems,
15(1):258–266, 2017.

[10] PhD Stud Alina-Elena LUPULESCU. Studies on stewart platform–a review.

[11] Ahmed F Mutlak and Amjad Jaleel Humaidi. A comparative study of synergetic and sliding
mode controllers for pendulum systems. Journal Européen des Systèmes Automatisés,
56(5):871, 2023.

[12] Yoshihiko Nakamura. Advanced robotics: redundancy and optimization. Addison-Wesley

Longman Publishing Co., Inc., 1990.

[13] Xingyu Qu, Zhenyang Li, Qilong Chen, Chengkun Peng, and Qinghe Wang. Research
on improved active disturbance rejection control strategy for hydraulic-driven stewart
stabilization platform. Industrial Robot: the international journal of robotics research
and application, 51(6):881–889, 2024.

[14] Zhengru Ren. A survey of modularized backstepping control design approaches to non-
linear ode systems. arXiv preprint arXiv:2305.02066, 2023.

[15] MK Safeena, KS Jiji, et al. Control of stewart platform using adaptive smooth integral
sliding mode algorithm. ISA transactions, 156:99–108, 2025.

[16] Diego Silva, Julio Garrido, and Enrique Riveiro. Stewart platform motion control au-
tomation with industrial resources to perform cycloidal and oceanic wave trajectories.
Machines, 10(8):711, 2022.

[17] Javier Velasco, Isidro Calvo, Oscar Barambones, Pablo Venegas, and Cristian Napole.
Experimental validation of a sliding mode control for a stewart platform used in aerospace
inspection applications. Mathematics, 8(11):2051, 2020.

[18] Ralph A Willoughby. Solutions of ill-posed problems (an tikhonov and vy arsenin). Siam
Review, 21(2):266, 1979.

[19] Fatemeh Zahedi and Zahra Zahedi. A review of neuro-fuzzy systems based on intelligent
control. Journal of Electrical and Electronic Engineering, 3(2-1):58–61, 2015.

[20] Jiangwei Zhao, Zhengjia Xu, Dongsu Wu, Yingrui Cao, and Jinpeng Xie. Six-dof stew-
art platform motion simulator control using switchable model predictive control. arXiv
preprint arXiv:2503.11300, 2025.

21